Early prediction of quality of service using interface-level metrics, code-level metrics, and antipatterns

Context: With the current high trends of deploying and using web services in practice, effective techniques for maintaining high quality of Service are becoming critical for both service providers and subscribers/users. Service providers want to predict the quality of service during early stages of development before releasing them to customers. Service clients consider the quality of service when selecting the best one satisfying their preferences in terms of price/budget and quality between the services offering the same features. The majority of existing studies for the prediction of quality of service are based on clustering algorithms to classify a set of services based on their collected quality attributes. Then, the user can select the best service based on his expectations both in terms of quality and features. However, this assumption requires the deployment of the services before being able to make the prediction and it can be time-consuming to collect the required data of running web services during a period of time. Furthermore, the clustering is only based on well-known quality attributes related to the services performance after deployment. Objective: In this paper, we start from the hypothesis that the quality of the source code and interface design can be used as indicators to predict the quality of service attributes without the need to deploy or run the services by the subscribers. Method: We collected training data of 707 web services and we used machine learning to generate association rules that predict the quality of service based on the interface and code quality metrics, and antipatterns. Results: The empirical validation of our prediction techniques shows that the generated association rules have strong support and high confidence which confirms our hypothesis that source code and interface quality metrics/antipatterns are correlated with web service quality attributes which are response time, availability, throughput, successability, reliability, compliance, best practices, latency, and documentation. Conclusion: To the best of our knowledge, this paper represents the first study to validate the correlation between interface metrics, source code metrics, antipatterns and quality of service. Another contribution of our work consists of generating association rules between the code/interface metrics and quality of service that can be used for prediction purposes before deploying new releases.